[FFmpeg-devel] [PATCH] Add DPX decoder rev-16

[Jimmy Christensen]

On 2009-06-03 16:27, Jimmy Christensen wrote:
> On 2009-06-03 12:28, Michael Niedermayer wrote:
>> On Wed, Jun 03, 2009 at 11:56:44AM +0200, Jimmy Christensen wrote:
>>> On 2009-06-03 11:33, Michael Niedermayer wrote:
>>>> On Wed, Jun 03, 2009 at 09:16:39AM +0200, Jimmy Christensen wrote:
>>>>> On 2009-06-02 22:21, Michael Niedermayer wrote:
>>>>>> On Tue, Jun 02, 2009 at 10:18:02AM +0200, Jimmy Christensen wrote:
>>>>>> [...]
>>>>>>> +typedef struct DPXContext {
>>>>>>> + AVFrame picture;
>>>>>>
>>>>>>> + int width;
>>>>>>> + int height;
>>>>>>
>>>>>
>>>>> Thanks. Will remove those and change the code accordingly.
>>>>>
>>>>>> redundant with AVCodecContext.width/...
>>>>>>
>>>>>> [...]
>>>>>>> + if(avctx->pix_fmt == PIX_FMT_RGB48BE) {
>>>>>>> + bytestream_put_be16(&dst,
>>>>>>> MAKE10to16(RED10(rgbBuffer)));
>>>>>>> + bytestream_put_be16(&dst,
>>>>>>> MAKE10to16(GREEN10(rgbBuffer)));
>>>>>>> + bytestream_put_be16(&dst,
>>>>>>> MAKE10to16(BLUE10(rgbBuffer)));
>>>>>>> + } else {
>>>>>>> + bytestream_put_le16(&dst,
>>>>>>> MAKE10to16(RED10(rgbBuffer)));
>>>>>>> + bytestream_put_le16(&dst,
>>>>>>> MAKE10to16(GREEN10(rgbBuffer)));
>>>>>>> + bytestream_put_le16(&dst,
>>>>>>> MAKE10to16(BLUE10(rgbBuffer)));
>>>>>>
>>>>>> this contains at least 6 unneeded shifts
>>>>>>
>>>>> Like this instead ?
>>>>>
>>>>> *dst++ = MAKE10to16(RED10(rgbBuffer));
>>>>> *dst++ = MAKE10to16(GREEN10(rgbBuffer));
>>>>> *dst++ = MAKE10to16(BLUE10(rgbBuffer));
>>>>
>>>> That is a good change as well
>>>>
>>>>
>>>>>
>>>>> or do you mean combining the MAKE10to16 with RED10? Personally I think
>>>>> that
>>>>> would make the code less readable.
>>>>
>>>> well you have to remove the unneeded operations, how you do that is
>>>> your
>>>> problem, droping all these macros is the cleanest way to write this
>>>> IMHO
>>>> but its up to you how you do it, it just must be optimal in terms of
>>>> speed
>>>>
>>>
>>> Here is the patch with the changes. I'm not good at working with
>>> bitshifting so I don't know how I can combine the bit shifting and
>>> masking
>>> in 1 go instead of 2.
>>
>> heres an example:
>>
>> RRRGGGBBB& 000111000 ->000GGG000
>> 000GGG000>> 2 -> 00000GGG0
>> 000GGG000>> 5 -> 00000000G
>> 00000GGG0 + 00000000G -> 00000GGGG
>>
>> does that help or should i fix the mask&shifts?
>>
>
> Thanks! That actually helped alot and I think I know what to do now.
>
> Hopefully this would be the last change needed to be accepted :)

Combined the macros.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ffmpegDPX-rev16.diff
Type: text/x-patch
Size: 8123 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20090603/c6e25126/attachment.bin>